CN113688326B

CN113688326B - Recommendation method, device, equipment and computer readable storage medium

Info

Publication number: CN113688326B
Application number: CN202111246496.7A
Authority: CN
Inventors: 钟子宏
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2021-10-26
Filing date: 2021-10-26
Publication date: 2022-02-08
Anticipated expiration: 2041-10-26
Also published as: CN113688326A

Abstract

The application provides a recommendation method, a recommendation device, recommendation equipment and a computer-readable storage medium; the method comprises the following steps: obtaining a plurality of user data corresponding to a plurality of trained recommendation models and a plurality of user identifications respectively corresponding to a plurality of marketing scene types, wherein the user data comprises user characteristic data and behavior data; inputting the user data into each recommendation model respectively to obtain a prediction vector corresponding to each marketing scene type; constructing a multi-scenario fusion matrix based on the prediction vectors corresponding to the marketing scenario types, wherein the multi-scenario fusion matrix comprises the prediction probabilities of a plurality of target user identifications under the marketing scenario types; determining a target marketing scene type corresponding to each target user identification based on the multi-scene fusion matrix; and when the recommendation opportunity is determined to be reached, sending recommendation information corresponding to the target marketing scene type to the terminal corresponding to each target user identification. Through the application, accurate recommendation can be realized.

Description

Recommendation method, device, equipment and computer readable storage medium

Technical Field

The present application relates to internet technologies, and in particular, to a recommendation method, apparatus, device, and computer-readable storage medium.

Background

A complete life cycle of a product generally includes: five periods of initial period, growth period, maturation period, decay period and withdrawal period. Corresponding to five life cycles, the intervention activities adopted by the enterprise are roughly divided into: initial stage-revivification, growth stage-revivification, loss early warning, maturity stage-loss early warning, loss recovery, payment revivification, payment activity and the like, decline stage-loss recovery, payment activity, payment loss recovery, quit stage-loss recovery, payment loss recovery and the like. Currently, for product life cycle management, enterprises generally adopt a single scene to perform marketing intervention of a single activity. However, many users exist in multiple life cycle stages of a product, so that one user can relate to marketing activities of multiple product life cycles, and the current life cycle marketing scheme cannot give the most advanced marketing campaign to the user, so that not only is the recommendation precision not high, but also the user is easy to be disturbed.

Disclosure of Invention

The embodiment of the application provides a recommendation method, a recommendation device and a computer-readable storage medium, which can improve recommendation precision.

The technical scheme of the embodiment of the application is realized as follows:

the embodiment of the application provides a recommendation method, which comprises the following steps:

obtaining a plurality of trained recommendation models respectively corresponding to a plurality of marketing scene types and a plurality of user characteristic data corresponding to a plurality of user identifications;

inputting the user characteristic data into each recommendation model respectively to obtain a prediction vector corresponding to each marketing scene type;

constructing a multi-scenario fusion matrix based on the prediction vectors corresponding to the marketing scenario types, wherein the multi-scenario fusion matrix comprises the prediction probabilities of a plurality of target user identifications under the marketing scenario types;

determining a target marketing scene type corresponding to each target user identification based on the multi-scene fusion matrix;

and when the recommendation opportunity is determined to be reached, sending recommendation information corresponding to the target marketing scene type to the terminal corresponding to each target user identification.

An embodiment of the present application provides a recommendation device, including:

the first acquisition module is used for acquiring the trained recommendation models corresponding to the marketing scene types and the user characteristic data corresponding to the user identifications;

the prediction module is used for respectively inputting the user characteristic data into each recommendation model to obtain a prediction vector corresponding to each marketing scene type;

the scene fusion module is used for constructing a multi-scene fusion matrix based on the prediction vectors corresponding to the marketing scene types, and the multi-scene fusion matrix comprises the prediction probabilities of a plurality of target user identifications under the marketing scene types;

the first determining module is used for determining a target marketing scene type corresponding to each target user identification based on the multi-scene fusion matrix;

and the first sending module is used for sending recommendation information corresponding to the target marketing scene type to the terminal corresponding to each target user identification when the recommendation opportunity is determined to be reached.

In some embodiments, the scene fusion module is further configured to:

determining a plurality of target user identifications corresponding to the marketing scene types based on the prediction vectors corresponding to the marketing scene types;

obtaining the prediction probability of each target user identification under at least one marketing scene type based on the prediction vector corresponding to each marketing scene type;

and constructing a multi-scenario fusion matrix based on the prediction probability of each target user identifier under at least one marketing scenario type.

In some embodiments, the scene fusion module is further configured to:

generating a probability matrix with missing values based on the prediction probability of each target user identification under at least one marketing scenario type;

wherein, the ith row and the jth column of the probability matrix are prediction probabilities of the ith target user identifier under the jth marketing scenario type, i =1,2, …, M, j =1,2, …, N, where M is a total number of target user identifiers and N is a total number of marketing scenario types, and when the pth target user identifier is not a target user under the qth marketing scenario type, the pth row and the qth column of the probability matrix are missing values;

and inputting the probability matrix into a trained collaborative filtering model for missing value prediction to obtain a multi-scene fusion matrix.

In some embodiments, the first determining module is further configured to:

determining scene sequencing vectors corresponding to the target user identifications based on the multi-scene fusion matrix, wherein the scene sequencing vectors are sequencing results of scene probability vectors corresponding to the target user identifications;

determining the highest scene probability of each target user identifier based on the scene sequencing vector corresponding to each target user identifier;

and determining the marketing scene type corresponding to the highest scene probability as a target marketing scene type.

In some embodiments, the apparatus further comprises:

the second determination module is used for determining the recommendation time when the preset recommendation interval duration is determined to be reached; or, the third determining module is configured to determine that a recommendation opportunity is reached when it is determined that the recommendation information is updated; or, the fourth determining module is configured to determine that the recommendation opportunity is reached when it is determined that the user logs in.

In some embodiments, the apparatus further comprises:

the second acquisition module is used for acquiring log data corresponding to each user identifier and determining training data based on the log data, wherein the training data comprises a plurality of training characteristic data and marketing scene labels corresponding to each training characteristic data;

the third acquisition module is used for acquiring training characteristic data corresponding to each marketing scene type and a preset recommendation model corresponding to each marketing scene type;

and the first training module is used for respectively training the preset recommendation models corresponding to the marketing scene types by using the training characteristic data corresponding to the marketing scene types to obtain the trained recommendation models corresponding to the marketing scene types.

In some embodiments, the second obtaining module is further configured to:

determining historical characteristic data and marketing scene labels corresponding to the user identifications based on the log data;

determining historical characteristic data corresponding to each marketing scene type based on the marketing scene labels;

and dividing the historical characteristic data corresponding to each marketing scene type to obtain training characteristic data corresponding to each marketing scene type and test data corresponding to each marketing scene type.

In some embodiments, the second obtaining module is further configured to:

determining identity characteristic data corresponding to each user identifier, consumption characteristic data and active characteristic data of each user identifier in a (K-1) th time period based on log data corresponding to each user identifier, wherein the (K-1) th time period is a last time period of a current time period, and K is an integer greater than 2;

based on log data corresponding to the user identification, if the fact that login is not performed before the Kth time period is determined, the marketing scene label of the user identification is determined to be a refresh scene;

if the user logs in the (K-1) th time period and does not log in the K th time period, determining that the marketing scene label of the user identifier is a loss early warning scene;

if the user is determined to log in the (K-2) th time period, the user is not logged in the (K-1) th time period, and the user is determined to log in the K th time period, and the marketing scene label of the user identifier is determined to be a loss retrieval scene;

if the fact that the user logs in but does not consume before the Kth time period is determined, logging in and consuming are carried out in the Kth time period, and the marketing scene label of the user identification is determined to be a newly added paid scene;

and if the marketing scene label of the user identification is determined to be the paid reflow scene, logging and consuming are carried out in the (K-2) th time period, logging and not consuming are carried out in the (K-1) th time period, logging and consuming are carried out in the Kth time period, and the marketing scene label of the user identification is determined to be the paid reflow scene.

In some embodiments, the first training module is further configured to:

respectively inputting the training characteristic data corresponding to each marketing scene type into a preset recommendation model corresponding to each marketing scene type for iterative training;

when the iteration ending condition is determined to be reached, obtaining each preliminarily trained recommendation model;

obtaining test data corresponding to each marketing scene type, wherein the test data comprises test characteristic data and test scene labels;

correspondingly inputting the test characteristic data corresponding to each marketing scene type into each preliminarily trained recommendation model to obtain prediction scene information corresponding to each recommendation model;

and when the condition of finishing training is determined to be reached based on the test scene labels and the prediction scene information corresponding to each recommendation model, determining each preliminarily trained recommendation model as each trained recommendation model.

In some embodiments, the apparatus further comprises:

the fourth obtaining module is used for obtaining new training data corresponding to the target recommendation model again when the target recommendation model which does not reach the training end condition is determined to exist based on the test scene label and the prediction scene information;

and the second training module is used for continuously training the target recommendation model by using the new training data until a training end condition is reached to obtain a trained recommendation model.

a memory for storing executable instructions;

and the processor is used for realizing the method provided by the embodiment of the application when executing the executable instructions stored in the memory.

The embodiment of the application provides a computer-readable storage medium, which stores executable instructions for causing a processor to execute the method for recommending provided by the embodiment of the application.

The embodiment of the present application provides a computer program product, which includes a computer program or instructions, and the computer program or instructions, when executed by a processor, implement the recommendation method provided by the embodiment of the present application.

The embodiment of the application has the following beneficial effects:

in the recommendation method provided by the embodiment of the application, firstly, trained recommendation models respectively corresponding to a plurality of marketing scene types and user characteristic data corresponding to a plurality of user identifications are obtained; then, inputting the user characteristic data into each recommendation model respectively to obtain a prediction vector corresponding to each marketing scene type; then, a multi-scene fusion matrix is constructed based on the prediction vectors corresponding to the marketing scene types, wherein the multi-scene fusion matrix comprises the prediction probabilities of a plurality of target user identifications under the marketing scene types; and then determining a target marketing scene type corresponding to each target user identification based on the multi-scene fusion matrix, and sending recommendation information corresponding to the target marketing scene type to a terminal corresponding to each target user identification when the recommendation opportunity is determined to be reached. Therefore, through marketing scene fusion, the optimal marketing scene type is determined for the user, the marketing recommendation precision is further improved, and the marketing recommendation cost can be reduced.

Drawings

FIG. 1 is a diagram of a marketing recommendation framework in the related art;

FIG. 2 is a schematic structural diagram of a recommendation system architecture provided in an embodiment of the present application;

fig. 3 is a schematic structural diagram of a server 400 provided in an embodiment of the present application;

FIG. 4 is a schematic flow chart of an implementation of a recommendation method provided in an embodiment of the present application;

FIG. 5A is a schematic diagram of a training process of a recommendation model provided in an embodiment of the present application;

FIG. 5B is a schematic diagram of another implementation of a training process of a recommendation model provided in an embodiment of the present application;

FIG. 6 is a schematic flowchart of still another implementation of the recommendation method according to an embodiment of the present application;

FIG. 7 is a schematic overall architecture diagram of a recommendation method provided in an embodiment of the present application;

fig. 8 is a schematic flow chart illustrating an implementation of a collaborative filtering-based multi-marketing scenario fusion recommendation method according to an embodiment of the present application;

FIG. 9 is a schematic diagram of a collaborative filtering matrix provided by an embodiment of the present application;

fig. 10 is a schematic diagram of a predicted score matrix provided in an embodiment of the present application.

Detailed Description

In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be described in further detail with reference to the attached drawings, the described embodiments should not be considered as limiting the present application, and all other embodiments obtained by a person of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict.

In the following description, references to the terms "first \ second \ third" are only to distinguish similar objects and do not denote a particular order, but rather the terms "first \ second \ third" are used to interchange specific orders or sequences, where appropriate, so as to enable the embodiments of the application described herein to be practiced in other than the order shown or described herein.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of the present application only and is not intended to be limiting of the application.

Before further detailed description of the embodiments of the present application, terms and expressions referred to in the embodiments of the present application will be described, and the terms and expressions referred to in the embodiments of the present application will be used for the following explanation.

1) Collaborative filtering is a widely used technique in recommendation systems. The technology predicts content that a user may be interested in and recommends this content to the user by analyzing similarities ("collaboration") between users or things;

2) scene fusion, a complex scene formed by fusing a plurality of service scenes, for example, a service scene in which an owner may be in a plurality of periods such as loss early warning, payment loss and the like at the same time in a life period of a product.

In order to better understand the recommendation method provided in the embodiments of the present application, a description is first made of a recommendation method in the related art.

The marketing recommendation scheme of the existing product life cycle management is mainly a single-scene marketing recommendation scheme, namely a classification prediction model under each scene is constructed by constructing scene labels (pull-in, loss early warning, loss recovery, new payment, loss payment, active payment and the like) of users at each stage of the product life cycle and characteristic data of the users, a machine learning model, a deep learning model and the like are trained by inputting sample data (the scene labels and the user characteristic data) to obtain a training model, a prediction sample is substituted into the training model to obtain the prediction model, and then a user ID and a classification score corresponding to the prediction positive sample are obtained, so that the users predicted as the positive sample are selected to recommend marketing activities.

However, the current technical solution belongs to a single-scene marketing recommendation scheme, and the marketing campaign needs to perform an operation campaign according to the scene where the user is located. Fig. 1 is a schematic diagram of a marketing recommendation framework in the related art, and as shown in fig. 1, a user group is judged in multiple scenes, and if a user has multiple marketing scenes at the same time, each marketing scene interferes with the user, so that user experience is seriously reduced.

In the related technology, for the situation involving a plurality of marketing scenes, a single scene independent marketing activity recommendation is adopted, and activities in each scene are independently carried out, however, a user may exist in a plurality of scenes, so that excessive waste of resources is caused, disturbance to the user is easily caused, and the experience of the user on the service marketing activity is reduced; because a user may appear in a plurality of operation scenes simultaneously, under multiple intervention, the effect of each marketing scene scheme and the effect of each model are difficult to distinguish, and a serious effect superposition problem exists, and in addition, the marketing recommendation scheme in the related art cannot determine which stage of five stages of the product life cycle the user is in, so that an optimal activity scheme cannot be selected for the product marketing activity.

Based on marketing recommendation schemes in the related technology, the method and the device apply the model to marketing scene fusion of product life cycle management, obtain target users and corresponding scene scores through the machine learning model, and then obtain probability scores of the users in each link scene of the product life cycle management by using the collaborative filtering model, so that an optimal scene marketing recommendation scheme is provided for the users, and recommendation precision is improved.

An exemplary application of the computer device provided in the embodiments of the present application is described below, and the computer device provided in the embodiments of the present application may be implemented as various types of user terminals such as a notebook computer, a tablet computer, a vehicle-mounted terminal, a desktop computer, a set-top box, a mobile device (e.g., a mobile phone, a portable music player, a personal digital assistant, a dedicated messaging device, and a portable game device), and may also be implemented as a server. In the following, an exemplary application will be explained when the device is implemented as a server.

Referring to fig. 2, fig. 2 is an architecture diagram of a recommendation system 100 provided in an embodiment of the present application, and as shown in fig. 2, the recommendation system includes a terminal 200 (in fig. 2, a terminal 200-1 and a terminal 200-2 are exemplarily shown), a network 300 and a server 400, where the terminal 200-1 and the terminal 200-2 are respectively connected to the server 400 through the network 300, and the network 300 may be a wide area network or a local area network, or a combination of both.

A variety of applications may be installed in the terminal 200, such as an instant messenger application, a third party payment application, a video viewing application, a shopping application, and the like. The terminal 200 can perform the instant messaging with relatives and friends or colleagues in life or work through the instant messaging application program, can realize the electronic payment through the third party payment application program, and can perform the online shopping through the shopping application program, etc. In the embodiment of the present application, the instant messaging application or the third party payment application may also be embedded with an applet that implements other business functions, for example, an online taxi taking applet, a take-away applet, a fueling applet, and the like.

For a product completing a certain business function, different pieces of recommendation information can be provided for different marketing scenes, and one user may correspond to a plurality of marketing scenes. In the embodiment of the application, after acquiring the trained recommendation models corresponding to the multiple marketing scene types and the user characteristic data corresponding to the multiple user identifications, the server 400 inputs the user characteristic data into each recommendation model respectively to obtain the prediction vector corresponding to each marketing scene type; then constructing a multi-scene fusion matrix based on the prediction vectors corresponding to the marketing scene types; and then determining a target marketing scene type corresponding to each target user identification based on the multi-scene fusion matrix, and sending recommendation information corresponding to the target marketing scene type to a user terminal corresponding to each target user identification when the recommendation opportunity is determined to be reached. Therefore, through marketing scene fusion, the optimal marketing scene type is determined for the user, the marketing recommendation precision is further improved, and the marketing recommendation cost can be reduced.

In some embodiments, the server 400 may be an independent physical server, may also be a server cluster or a distributed system formed by a plurality of physical servers, and may also be a cloud server providing basic cloud computing services such as cloud services, a cloud database, cloud computing, a cloud function, cloud storage, a network service, cloud communication, middleware services, a domain name service, a security service, smart transportation, an intelligent internet automobile, a CDN, and a big data and artificial intelligence platform.

Referring to fig. 3, fig. 3 is a schematic structural diagram of a server 400 according to an embodiment of the present application, where the server 400 shown in fig. 3 includes: at least one processor 410, memory 450, at least one network interface 420, and a user interface 430. The various components in server 400 are coupled together by a bus system 440. It is understood that the bus system 440 is used to enable communications among the components. The bus system 440 includes a power bus, a control bus, and a status signal bus in addition to a data bus. For clarity of illustration, however, the various buses are labeled as bus system 440 in FIG. 3.

The Processor 410 may be an integrated circuit chip having Signal processing capabilities, such as a general purpose Processor, a Digital Signal Processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like, wherein the general purpose Processor may be a microprocessor or any conventional Processor, or the like.

The user interface 430 includes one or more output devices 431, including one or more speakers and/or one or more visual displays, that enable the presentation of media content. The user interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid state memory, hard disk drives, optical disk drives, and the like. Memory 450 optionally includes one or more storage devices physically located remote from processor 410.

The memory 450 includes either volatile memory or nonvolatile memory, and may include both volatile and nonvolatile memory. The nonvolatile Memory may be a Read Only Memory (ROM), and the volatile Memory may be a Random Access Memory (RAM). The memory 450 described in embodiments herein is intended to comprise any suitable type of memory.

In some embodiments, memory 450 is capable of storing data, examples of which include programs, modules, and data structures, or a subset or superset thereof, to support various operations, as exemplified below.

An operating system 451, including system programs for handling various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and handling hardware-based tasks;

a network communication module 452 for communicating to other computing devices via one or more (wired or wireless) network interfaces 420, exemplary network interfaces 420 including: bluetooth, wireless compatibility authentication (WiFi), and Universal Serial Bus (USB), etc.;

an input processing module 453 for detecting one or more user inputs or interactions from one of the one or more input devices 432 and translating the detected inputs or interactions.

In some embodiments, the apparatus provided by the embodiments of the present application may be implemented in software, and fig. 3 illustrates a recommendation apparatus 454 stored in a memory 450, which may be software in the form of programs and plug-ins, and the like, and includes the following software modules: a first obtaining module 4541, a predicting module 4542, a scene fusing module 4543, a first determining module 4544 and a first transmitting module 4545, which are logical and thus may be arbitrarily combined or further divided according to the functions implemented. The functions of the respective modules will be explained below.

In other embodiments, the apparatus provided in the embodiments of the present Application may be implemented in hardware, and for example, the apparatus provided in the embodiments of the present Application may be a processor in the form of a hardware decoding processor, which is programmed to execute the recommended method provided in the embodiments of the present Application, for example, the processor in the form of the hardware decoding processor may be one or more Application Specific Integrated Circuits (ASICs), DSPs, Programmable Logic Devices (PLDs), Complex Programmable Logic Devices (CPLDs), Field Programmable Gate Arrays (FPGAs), or other electronic components.

The recommendation method provided by the embodiment of the present application will be described in conjunction with exemplary applications and implementations of the server provided by the embodiment of the present application.

Referring to fig. 4, fig. 4 is a schematic flow chart of an implementation of the recommendation method provided in the embodiment of the present application, which will be described with reference to the steps shown in fig. 4.

Step S101, obtaining a plurality of user characteristic data corresponding to a plurality of trained recommendation models and a plurality of user identifications respectively corresponding to a plurality of marketing scene types.

In the embodiment of the application, the marketing scenario types may include a pull-up scenario type, a loss early warning scenario type, a loss pull-up scenario type, a pay reflow scenario type, and the like. Different trained recommendation models correspond to different marketing scene types, and the recommendation models can be neural network models, such as convolutional neural network models or deep learning neural network models. The different trained recommended models may have the same model structure but different model parameters, or may have different model structures but different model parameters.

The user identification may be a registration ID of the user, which is unique, unlike an account number. User characteristic data may include identity characteristic data, consumption characteristic data, active characteristic data, and the like. The identity data may include: the consumption characteristic data can comprise recharging amount, consumption amount, recharging times, recharging days, interval of the first recharging from the current time days and the like; the active feature data includes: the number of active days, the active duration, the number of active functions, the interval of the registration time from the current time day.

And S102, respectively inputting the user characteristic data into each trained recommendation model to obtain a prediction vector corresponding to each marketing scene type.

In the embodiment of the application, the user characteristic data corresponding to the plurality of user identifications are respectively input into each trained recommendation model to predict the scenes corresponding to the user identifications, so that the prediction vectors corresponding to the marketing scene types are obtained. The dimension of the prediction vector is the same as the number of the users, for example, if there are 100 users, the prediction vector is 1 × 100, and each prediction value in the prediction vector corresponding to a certain marketing scenario type represents the probability that each user is the marketing scenario.

And S103, constructing a multi-scene fusion matrix based on the prediction vectors corresponding to the marketing scene types.

The multi-scenario fusion matrix comprises prediction probabilities of a plurality of target user identifications under each marketing scenario type, wherein the prediction probabilities can be prediction values in prediction vectors or predicted based on a collaborative filtering algorithm.

When the method is implemented, target user identifications of various marketing scene types can be determined based on prediction vectors corresponding to the various marketing scene types, then probability matrixes with missing values are constructed based on the target user identifications and prediction values of the various target user identifications in the prediction vectors, and then the missing values in the probability matrixes are predicted by using trained collaborative filtering models to obtain multi-scene fusion matrixes.

And step S104, determining the target marketing scene type corresponding to each target user identification based on the multi-scene fusion matrix.

Each row of the multi-scenario fusion matrix may be a probability that a target user identifier belongs to each marketing scenario type. When the step is implemented, the marketing scenario type corresponding to the highest value in the probability that the target user identifier is subordinate to each marketing scenario type can be determined as the target marketing scenario type of the target user identifier.

For example, there are five marketing scene types, which are respectively a pull-up scene type, a loss early warning scene type, a loss recovery scene type, and a pay-for-pull-up scene type, probabilities of a certain target user identifier in the five marketing scene types are respectively 0.6, 0.3, 0.2, 0.3, and 0.2, and since 0.6 is the highest value, the pull-up scene type corresponding to 0.6 is determined as the target marketing scene type of the target user identifier.

And step S105, when the recommendation opportunity is determined to be reached, sending recommendation information corresponding to the target marketing scene type to the terminal corresponding to each target user identification.

In the implementation, the recommendation time may be determined to be reached when it is determined that new recommendation information is issued, that is, when it is determined that the recommendation information is updated or when a preset recommendation interval is reached, or when it is determined that the user logs in.

Because different marketing scene types correspond to different recommendation information, for example, for a pull-up scene type, the recommendation information may be first order discount information, and for a loss early warning scene type, the recommendation information may be consumption discount information, and may also be recommendation friend discount information, and the like.

When the step S105 is implemented, the server may send the application program or the applet to the user after monitoring that the user opens the application program, or may push the application program or the applet to the user in a short message manner.

In the recommendation method provided by the embodiment of the application, when the recommendation opportunity is determined to be reached, firstly, trained recommendation models respectively corresponding to a plurality of marketing scene types and user characteristic data corresponding to a plurality of user identifications are obtained; then, inputting the user characteristic data into each recommendation model respectively to obtain a prediction vector corresponding to each marketing scene type; then, a multi-scene fusion matrix is constructed based on the prediction vectors corresponding to the marketing scene types, wherein the multi-scene fusion matrix comprises the prediction probabilities of a plurality of target user identifications under the marketing scene types; and then determining a target marketing scene type corresponding to each target user identification based on the multi-scene fusion matrix, and sending recommendation information corresponding to the target marketing scene type to a terminal corresponding to each target user identification. Therefore, through marketing scene fusion, the optimal marketing scene type is determined for the user, the marketing recommendation precision is further improved, and the marketing recommendation cost can be reduced.

In some embodiments, the step S103 "building a multi-scenario fusion matrix based on the prediction vectors corresponding to the marketing scenario types" may be implemented by:

and step S1031, determining a plurality of target user identifications corresponding to each marketing scene type based on the prediction vector corresponding to each marketing scene type.

When the step is implemented, the user identifier corresponding to the predicted value larger than the preset threshold value in the prediction vector corresponding to each marketing scenario type may be determined as the target user identifier. For example, if the preset threshold is 0.5, the user identifiers with the predicted values larger than 0.5 are sequentially selected from the prediction vectors corresponding to the marketing scene types, that is, the target user identifiers corresponding to the marketing scene types are obtained.

The multiple target user identifications contained in different marketing scenario types may be partially or completely identical.

Step S1032, obtaining the prediction probability of each target user identification under at least one marketing scene type based on the prediction vector corresponding to each marketing scene type.

The prediction probability obtained in this step is a prediction value in the prediction vector, and the prediction probability is greater than a preset threshold. For example, the user 1 is only a target user of the pull-up scene type, the prediction probability in the pull-up scene type is 0.8, and the user 2 is a target user of two marketing scene types, namely, the churn early warning and the pay-for-pull-up, then the prediction probability corresponding to the user 2 is obtained from the prediction vector corresponding to the churn early warning scene type, the prediction probability is assumed to be 0.65, and the prediction probability corresponding to the user 2 is obtained from the prediction vector corresponding to the pay-for-pull-up scene type, and the prediction probability is assumed to be 0.8. The user 3 is a target user for early warning of loss, recovery of loss and payment and backflow, and the corresponding prediction probabilities are 0.75, 0.83 and 0.87 respectively.

Step S1033, a multi-scenario fusion matrix is constructed based on the prediction probability of each target user identification under at least one marketing scenario type.

In the embodiment of the application, when the step is implemented, a probability matrix with missing values is generated based on the prediction probability of each target user identifier under at least one marketing scenario type; the method comprises the steps that the ith row and the jth column of a probability matrix are prediction probabilities of ith target user identifications under the jth marketing scenario type, i =1,2, …, M, j =1,2, … and N are provided, wherein M is the total number of the target user identifications, N is the total number of the marketing scenario types, and when the pth target user identifications are not target users under the pth marketing scenario type, the pth row and the pth column of the probability matrix are missing values. And then inputting the probability matrix into a trained collaborative filtering model for missing value prediction to obtain a multi-scene fusion matrix.

Taking the above example, assuming that the probability matrix is constructed, the ranking of the marketing scenario types is: the target users comprise a user 1, a user 2 and a user 3, and at this time, based on the probability value information, the following probability matrix can be constructed:

since the user 1 is only the target user of the pull-up scene, the 2 nd column to the 5 th column in the 1 st row are all missing values, and 0 represents the missing value in the probability matrix, but actually there is no value, since the user 2 is the target user of the two marketing scene types of the churn warning and the pay-up pull-up, the first column, the third column and the fourth column in the 2 nd row are missing values, and since the user 3 is the target user of the churn warning, the churn recovery and the pay-back, the 1 st column and the 5 th column in the 3 rd row are missing values.

And then inputting the probability matrix into a trained collaborative filtering model, and predicting missing values in the probability matrix to obtain a multi-scene fusion matrix, namely the multi-scene fusion matrix has no missing values, and the missing values are predicted based on probability values of users in other scenes.

Through the steps S1031 to S1033, the probability matrix is constructed through the predicted values of the target user identifications in the prediction vectors corresponding to the marketing scene types, so that the missing values in the probability matrix are predicted by utilizing the collaborative filtering matrix, and the multi-scene fusion matrix is obtained, so that a data basis is provided for subsequently determining the marketing recommendation information of the user, and the excessive waste of the recommendation resources when the user is in a plurality of marketing scenes can be avoided.

In some embodiments, the step S104 "determining the target marketing scenario type corresponding to each target user identifier based on the multi-scenario fusion matrix" may be implemented by the following steps S1041 to S1043, which are described below.

Step S1041, determining a scene ordering vector corresponding to each target user identifier based on the multi-scene fusion matrix.

The scene ranking vector is a ranking result of the scene probability vector corresponding to the target user identifier, and the ranking result can be obtained by ranking from large to small or from small to large. For example, in the multi-scene fusion matrix, the scene probability vectors corresponding to the user 2 are 0.35, 0.65, 0.23, 0.47, and 0.8, which are sorted from large to small, and then the scene sorting vectors corresponding to the user 2 are 0.8, 0.65, 0.47, 0.35, and 0.23.

Step S1042, determining the highest scene probability of each target user id based on the scene ranking vector corresponding to each target user id.

In implementation, if the scene sequencing vectors are arranged in the order from small to large, in this step, the last probability value in the scene sequencing vectors is obtained, so as to obtain the highest scene probability, and if the scene sequencing vectors are arranged in the order from large to small, in this step, the first probability value in the scene sequencing vectors is obtained, so as to obtain the highest scene probability. Taking the above example, the highest scene probability for user 2 is 0.8.

And S1043, determining the marketing scene type corresponding to the highest scene probability as a target marketing scene type.

After the highest scene probability is obtained, the marketing scene type corresponding to the highest scene probability can be obtained correspondingly. Taking advantage of the above example, the marketing scenario type corresponding to the highest scenario probability of 0.8 in the user 2 is the pay-for-update scenario type.

Through the steps S1041 to S1043, after the multi-scene fusion matrix is obtained, the scene ordering vector of each target user identifier can be determined, and the target marketing scene type of each target user identifier is determined based on the scene ordering vector, so that even if the user is in a plurality of marketing scenes, an optimal target marketing scene type can be selected from the marketing scenes, thereby providing more accurate and effective recommendation information for the user.

In the implementation of the present application, before step S101, trained recommendation models corresponding to various marketing scenario types can be obtained through the steps shown in fig. 5A, which is described below with reference to fig. 5A.

And S001, acquiring log data corresponding to each user identifier, and determining training data based on the log data.

The log data corresponding to each user identifier may be log data within a preset time length, for example, the log data may be log data of a week before the current time, or may also be log data of five days before the current time, and the log data is based on data generated by logging in and using a corresponding application program or applet within the preset time length by each user identifier.

In the embodiment of the application, after the log data are obtained, data processing is performed on the log data to obtain historical feature data and marketing scenario labels of each user identifier, and then the marketing scenario labels of the historical feature data corresponding to each user identifier are divided into training feature data and testing feature data. The training data comprises a plurality of training characteristic data and marketing scenario labels corresponding to the training characteristic data.

And S002, acquiring training characteristic data corresponding to each marketing scene type and a preset recommendation model corresponding to each marketing scene type.

In the step, the marketing scene labels are utilized to divide the training characteristic data to obtain the training characteristic data corresponding to each marketing scene type.

The preset recommendation models corresponding to the marketing scene types are initial untrained recommendation models, and each preset recommendation model can be the same model structure, the same initial parameters, or different model structures and different initial parameters, and is not limited in the embodiment of the application.

And S003, training the preset recommendation models corresponding to the marketing scene types respectively by using the training feature data corresponding to the marketing scene types to obtain the trained recommendation models corresponding to the marketing scene types.

In the step, training characteristic data corresponding to each marketing scene type is input into a corresponding preset recommendation model to obtain a prediction result of each user identification, and then back propagation training is performed on the preset recommendation model by using the marketing scene labels of the user identifications and the prediction results, so that parameters of the preset recommendation model are adjusted until a preset training end condition is reached, and at the moment, a preliminarily trained recommendation model is obtained. In some embodiments, a preset recommendation model corresponding to each marketing scenario type may also be iteratively trained by using a gradient descent method, and when an iteration end condition is reached, a track is preliminarily trained. And then, testing the prediction performance of the preliminarily trained recommendation model by using the test data until the recommendation model meeting the preset index is determined, and obtaining the finally trained recommendation model at the moment.

In some embodiments, the "determining training data based on the log data" in the above step S001 may be implemented by:

and S0011, determining historical characteristic data and marketing scene labels corresponding to the user identifications on the basis of the log data.

This step S0011 can be implemented by:

step S0111, based on the log data corresponding to each user identifier, identity characteristic data corresponding to each user identifier, consumption characteristic data and active characteristic data of each user identifier in the (K-1) th time period are determined.

In the embodiment of the present application, the obtained log data may include all log data registered from the user, or may be data of multiple time periods, for example, at least three time periods, five time periods, or ten time periods.

One of the time periods may be a preset time duration, for example, 24 hours, and may be 12 hours, three days, five days, and the like. Assuming that the current time period is the kth period, and K is an integer greater than 2, the (K-1) th time period is a previous time period of the current time period. The identity data corresponding to the user identifier may include the user identifier, gender, age, location area, and the like. The consumption profile data at the (K-1) th time period may include: the recharging amount, the consumption amount, the recharging times, the recharging days, the interval between the first recharging and the current time days and the like in the (K-1) th time period; the active signature data for the (K-1) th time period includes: the number of active days, the active time length, the number of active functions and the interval between the registration time and the current time day in the (K-1) th time period.

Step S0112, based on log data corresponding to the user identification, if it is determined that login is not performed before the Kth time period, it is determined that the marketing scene label of the user identification is a refresh scene.

In this step, if the user logs in before the kth time period, it indicates that the user has never used the service functions of the application and is a new user, and therefore, it is determined that the marketing scenario label identified by the user is a pull-up scenario at this time.

And S0113, if the user logs in the (K-1) th time period and does not log in the K time period, determining that the marketing scene label of the user identifier is a loss early warning scene.

If the user logs in the (K-1) th time period and does not log in the K th time period, the fact that the user is likely to have the loss risk is indicated, and therefore the marketing scenario label identified by the user is determined as the loss early warning scenario.

And S0114, if the user logs in the (K-2) th time period and does not log in the (K-1) th time period, the user logs in the K th time period, and the marketing scene label of the user identifier is determined to be a loss recovery scene.

If the user logs in the (K-2) th time period and does not log in the (K-1) th time period, and logs in again in the K th time period, the fact that the user may lose but does not lose is indicated, and therefore the marketing scenario label identified by the user is determined to be a loss recovery scenario.

And S0115, if the fact that the user logs in but does not consume before the Kth time period is determined, logging in and consuming are performed in the Kth time period, and the marketing scene label of the user identification is determined to be a new pay scene.

And S0116, if the fact that the user logs in and consumes in the (K-2) th time period is determined, the fact that the user logs in and does not consume in the (K-1) th time period is determined, the fact that the user logs in and consumes in the K th time period is determined, and the marketing scene label marked by the user is determined to be a paid reflow scene.

And S0012, determining historical characteristic data corresponding to each marketing scene type based on the marketing scene labels.

In the embodiment of the application, historical feature data corresponding to each user identifier can be firstly divided into sparse features and dense features, wherein the sparse features can be ID (identity) features and ID (identity) features, and the sparse features are subjected to one-hot (onehot) processing at the moment, so that the identification degree can be increased by marking the positions of the features; the dense feature is mainly a continuous numerical feature, and is generally subjected to Principal Component Analysis (PCA) decorrelation processing, normalization (normalization) processing, feature discretization processing, and the like, so that the influence of dimension can be eliminated, and the identification degree of the model can be improved.

After the sparse characteristic and the dense characteristic are correspondingly processed, the processed historical characteristic data are classified according to the marketing scene labels, and therefore historical characteristic data corresponding to various marketing scene types are obtained.

And S0013, dividing the historical characteristic data corresponding to each marketing scene type to obtain training characteristic data corresponding to each marketing scene type and test data corresponding to each marketing scene type.

When the step is realized, the historical characteristic data corresponding to each marketing scene type can be randomly segmented according to a certain proportion, so that training characteristic data corresponding to each marketing scene type and test data corresponding to each marketing scene type are obtained. For example, the training data and the test data may be randomly divided according to a ratio of 8:2, so as to obtain training feature data and test data corresponding to each marketing scenario.

The training characteristic data is used for training the recommendation model, the test data is used for evaluating the preliminarily trained recommendation model to determine whether the preliminarily trained recommendation model meets the evaluation standard, and if not, the training is continued, so that the finally trained recommendation model meets the evaluation standard, and the accuracy of the prediction result of the finally trained recommendation model is ensured.

In some embodiments, the step S003 "training the preset recommendation models corresponding to the marketing scene types respectively by using the training feature data corresponding to the marketing scene types to obtain the trained recommendation models corresponding to the marketing scene types" may be implemented by steps S0031 to S0039 shown in fig. 5B, and each step is described below with reference to fig. 5B.

And S0031, respectively inputting the training characteristic data corresponding to each marketing scene type into a preset recommendation model corresponding to each marketing scene type for iterative training.

Step S0032, it is determined whether an iteration end condition is reached.

Here, the iteration end condition may be that a preset number of iterations is reached, or that a minimum value of the objective function is reached. If it is determined that the iteration end condition is reached, the routine proceeds to step S0033, and if it is determined that the training end condition is not reached, the routine proceeds to step S0031 to continue training.

And step S0033, obtaining each preliminarily trained recommendation model.

And S0034, obtaining test data corresponding to each marketing scene type.

The test data comprises test characteristic data and a test scene label.

And S0035, correspondingly inputting the test characteristic data corresponding to each marketing scene type into each preliminarily trained recommendation model to obtain the prediction scene information corresponding to each recommendation model.

And S0036, determining whether training ending conditions are met or not based on the test scene labels and the prediction scene information corresponding to each recommendation model.

Here, when the step S0036 is implemented, the evaluation index value of each recommendation model may be determined based on the test scenario label and the prediction scenario information corresponding to each recommendation model, where the evaluation index may include at least one of a recall ratio, a precision ratio, and an Area Under the Curve (AUC), and then whether the evaluation index value of each recommendation model reaches a preset index threshold is respectively determined, and when the evaluation index value of a certain recommendation model reaches the index threshold, it is determined that the preliminarily trained recommendation model reaches the training end condition, at this time, step S0037 is performed; when there is a certain recommended model whose evaluation index value does not reach the index threshold, it is determined that there is a target recommended model that does not reach the training end condition, and the process proceeds to step S0038.

And S0037, determining the preliminarily trained recommendation models as the trained recommendation models.

And step S0038, acquiring new training data corresponding to the target recommendation model again.

The target recommendation model is also the recommendation model which does not reach the training end condition. When the step is realized, training data of marketing scene types corresponding to the marketing scene labels as the target recommendation models are obtained.

And S0039, continuing training the target recommendation model by using the new training data until a training end condition is reached, and obtaining the trained target recommendation model.

And step S0039, when the training is realized, training the preliminarily trained target recommendation model by using the training characteristic data in the new training data until the iteration end condition is reached, then obtaining the test data again, performing index evaluation on the target recommendation model to determine whether the training end condition is reached, when the training end condition is reached, obtaining the trained target recommendation model, and when the training end condition is not reached, continuously repeating the step S0038 and the step S0039 until the trained target recommendation model is obtained.

Based on the foregoing embodiments, a recommendation method is further provided in an embodiment of the present application, and is applied to the network architecture shown in fig. 2, and fig. 6 is a schematic view of a further implementation flow of the recommendation method provided in the embodiment of the present application. The respective steps will be described below with reference to fig. 6.

Step S601, the server obtains a plurality of user characteristic data corresponding to a plurality of trained recommendation models and a plurality of user identifications respectively corresponding to a plurality of marketing scene types.

Here, the plurality of user ids may be all user ids registered in the travel service application, or may be user ids currently in a login state. The user characteristic data may include identity characteristic data, consumption characteristic data of the current time period, active characteristic data of the current time period, and in some embodiments, the consumption characteristic data may further include information of a type (quantity, times, value) of gift bags/gift certificates received by the user, a type (quantity, value) of gift bags/gift certificates used, a type (quantity, value) of gift bags/gift certificates expired, and the like.

Step S602, the server inputs the user characteristic data into each trained recommendation model respectively to obtain a prediction vector corresponding to each marketing scene type.

Here, different trained recommendation models correspond to different marketing scenario types, and the recommendation model may be a neural network model, for example, a convolutional neural network model, or a deep learning neural network model. The different trained recommended models may have the same model structure but different model parameters, or may have different model structures but different model parameters.

Step S603, the server determines a plurality of target user identifiers corresponding to each marketing scenario type based on the prediction vector corresponding to each marketing scenario type.

During implementation, the user identifier with the predicted value larger than the preset threshold value in the prediction vector corresponding to each marketing scene type is determined as the target user identifier corresponding to the impact scene type. The set of target user identifications for different marketing scenario types typically differ.

Step S604, the server obtains the prediction probability of each target user identifier under at least one marketing scenario type based on the prediction vector corresponding to each marketing scenario type.

Step S605, the server constructs a multi-scenario fusion matrix based on the prediction probability of each target user identification under at least one marketing scenario type.

When the step is realized, a probability matrix is generated based on the prediction probability of each target user identification under at least one marketing scenario type. In the actual implementation process, because not every target user identifier exists in all marketing scenarios, the probability matrix is provided with missing values, and in order to ensure the accuracy of the prediction result, after the probability matrix is obtained, the probability matrix can be input into a trained collaborative filtering model for missing value prediction, so that a multi-scenario fusion matrix is obtained.

In the embodiment of the present application, the trained collaborative filtering model may be a neural network model, for example, a deep learning neural network model, a cyclic neural network model, or the like.

Step S606, the server determines the scene ordering vector corresponding to each target user identifier based on the multi-scene fusion matrix.

The scene ranking vector is a ranking result of the scene probability vector corresponding to the target user identifier, and the ranking result may be a result obtained by ranking according to a descending order or a result obtained by ranking according to a descending order.

Step S607, the server determines the highest scene probability of each target user identifier based on the scene ranking vector corresponding to each target user identifier.

After the scene sequencing vector corresponding to each target user identifier is obtained, the highest scene probability of each target user identifier can be obtained.

Step S608, the server determines the marketing scenario type corresponding to the highest scenario probability of each target user identifier as the target marketing scenario type corresponding to each target user identifier.

In step S609, the terminal starts the travel service application in response to the received start instruction.

The starting instruction can be triggered by the user clicking an icon of the trip service application program, can also be triggered by the user making a gesture for starting the trip service application program, and can also be triggered based on voice sent by the user for starting the trip service application program.

Step S610, when monitoring that the terminal starts the application and logs in, the server obtains a user identifier corresponding to the terminal.

The user identifier corresponding to the terminal may be allocated by the server when the user registers the travel service application program, and the user identifier may be a phone number of the user, a mailbox address of the user, an account identifier of instant messaging of the user, and the like.

Step S611, when determining that the user identifier corresponding to the terminal is the target user identifier, the server determines that the recommendation opportunity is reached.

In this step, the server matches the user identifier corresponding to the terminal with a plurality of target user identifiers, and if the matching is successful, the user identifier corresponding to the terminal is determined to be the target user identifier, and at this time, it is determined that the recommendation opportunity is reached.

Step S612, the server sends recommendation information corresponding to the target marketing scene type to the terminal.

Step S613, the terminal presents the received recommendation information.

When the method is implemented, the terminal can load a floating layer on a display interface of the terminal and display the received recommendation information on the floating layer. In the embodiment of the application, the recommendation information may be coupon information, and may also be invitation friends to receive the coupon information.

And step S614, when the terminal responds to the coupon information provided by the recommendation information, the coupon information held by the terminal is updated.

In the recommendation method provided by the embodiment of the application, a server acquires trained recommendation models respectively corresponding to a plurality of marketing scene types and user characteristic data corresponding to a plurality of user identifications; then, inputting the user characteristic data into each recommendation model respectively to obtain a prediction vector corresponding to each marketing scene type; determining a plurality of target user identifications corresponding to the marketing scene types based on the prediction vectors corresponding to the marketing scene types; obtaining the prediction probability of each target user identification under at least one marketing scene type based on the prediction vector corresponding to each marketing scene type; then constructing a multi-scenario fusion matrix based on the prediction probability of each target user identifier under at least one marketing scenario type; and then determining the target marketing scene type corresponding to each target user identification based on the multi-scene fusion matrix, determining whether the user identification corresponding to the terminal is the target user identification when determining that the user terminal starts and logs in a travel service application program, if so, determining that the recommendation opportunity is reached, at the moment, sending recommendation information corresponding to the target marketing scene type to the terminal by the server, and presenting the recommendation information after the terminal receives the recommendation information. Therefore, through marketing scene fusion, the optimal marketing scene type is determined for the user, the marketing recommendation accuracy is further improved, frequent disturbance to the user can be reduced, and the marketing recommendation cost can be reduced.

Next, an exemplary application of the embodiment of the present application in a practical application scenario will be described.

In the embodiment of the present application, a digital marketing scenario of a vehicle enterprise is taken as an example for explanation. The digital marketing scenario of the vehicle-enterprise can comprise a taxi taking service, a fuel filling service and the like. The preferential fueling service is a typical case of the digital marketing of the vehicle enterprises, and the service product module can be embedded in an instant messaging application program or an applet in a digital payment application program, and can also be a separate application program.

The vehicle owner can relate to the initial stage, the growth stage, the maturation stage, the decline stage and the quit stage of the life cycle of the vehicle owner in the preferential fueling product. Therefore, the enterprise carries out marketing scheme activities such as a pull-up activity, a loss early warning activity, a loss recovery activity, a payment pull-up activity, a payment backflow activity and the like on the vehicle owner through the preferential fueling service module. If each vehicle owner is recommended independently for various marketing activities, the marketing cost can be greatly increased, and disturbance intervention is carried out on the vehicle owners for many times, so that the vehicle owners are easily disliked, and the experience of the vehicle owners on travel products is reduced.

Fig. 7 is an overall architecture schematic diagram of the recommendation method provided in the embodiment of the present application, and since there are a plurality of marketing scenarios corresponding to the car owners in each car owner group, as shown in fig. 7, the car owners in each car owner group are subjected to marketing scenario fusion, classification prediction scores of the car owners in respective scenario marketing scenarios are determined, a collaborative filtering score matrix with missing values of the car owners and the scenario operation scenarios is constructed, and a collaborative filtering method is used to fill probability values of the marketing scenarios missing of the car owners, so as to obtain the score matrix. In the embodiment of the application, the marketing schemes of all vehicle owners in the vehicle main group to all marketing scenes are sequenced, so that the priority of the life cycles of products where all vehicle owners are located in all the vehicle main groups is obtained, and corresponding marketing strategies are carried out on the vehicle owners according to the priority of the marketing schemes.

It should be noted that, by taking 701 in fig. 7 as an example, only the marketing scene ranking of the first vehicle owner in the vehicle owner group is exemplarily shown in fig. 7, in the embodiment of the present application, the marketing scene ranking of each vehicle owner is determined, so that the vehicle owner is correspondingly recommended according to the ranking result.

In the embodiment of the application, a marketing scene selection scheme of collaborative filtering is adopted, scenes are scored and sorted through a collaborative filtering method, and activity intervention is performed on the marketing scene with the highest score each time, so that the cost can be effectively reduced, the marketing activity target is reached, meanwhile, the disturbance of marketing activities on car owners is reduced, and the product experience of the car owners is improved.

Fig. 8 is a schematic flow chart of an implementation of the collaborative filtering-based multi-marketing scenario fusion recommendation method provided in the embodiment of the present application, and as shown in fig. 8, the flow chart includes the following nine stages: a data processing stage 801, a single scene model training and testing stage 802, a single scene prediction scoring stage 803, a single scene vehicle main classification stage 804, a multi-scene fusion matrix construction stage 805, a Collaborative Filtering multi-scene (CF) model training stage 806, a model prediction scoring stage 807, a multi-scene score sorting stage 808, and a scene recommendation stage 809. The implementation of each stage is explained below.

In the data processing stage 801, vehicle owner log data are first obtained, and then the vehicle owner log data are respectively processed into vehicle owner scene labels and vehicle owner characteristic data, so that training samples, test samples and prediction samples in each scene are constructed based on the vehicle owner scene labels and the vehicle owner characteristic data.

Wherein, car owner scene label includes: a refresh scene, a loss early warning scene, a loss recovery scene, a payment refresh scene, a payment backflow scene, etc., wherein:

the pull-new scene tag is constructed as follows: before the K period, the vehicle owner never logs in the preferential refueling function module, and the K period logs in the module to indicate that the vehicle owner is a newly increased vehicle owner of the preferential refueling module and is marked as 1; otherwise, it is marked 0.

The loss early warning scene label is constructed as follows: the vehicle owner logs in the preferential fueling function module in the K-1 period, and the vehicle owner is represented as a lost vehicle owner in the preferential fueling module in the K period and marked as 1 when the vehicle owner does not log in the preferential fueling function module in the K period; otherwise, the vehicle owner logs in the preferential fueling function module in the K-1 period, and also logs in the module in the K period, which indicates that the vehicle owner is a reserved vehicle owner in the preferential fueling module in the K period and is marked as 0.

The loss retrieval scene label is constructed as follows: the vehicle owner logs in the preferential refueling function module in the K-2 period, does not log in the module in the K-1 period, logs in the module in the K period, shows that the vehicle owner is a reflow vehicle owner in the preferential refueling module in the K period, and is marked as 1; otherwise, the vehicle owner logs in the preferential fueling function module in the K-2 period, does not log in the module in the K-1 period, does not log in the module in the K period, shows that the vehicle owner is a lost vehicle owner in the preferential fueling module in the K period, and is marked as 0.

The pay pull new scenario label is constructed as follows: the vehicle owner logs in the preferential refueling function module before the K period but does not pay the order, and the vehicle owner logs in the preferential refueling function module and has a payment order in the K period, which indicates that the vehicle owner is a new paid vehicle owner of the preferential refueling module and is marked as 1; otherwise, it is marked 0.

The pay reflow scenario label is constructed as follows: the vehicle owner logs in the preferential refueling function module in the K-2 period and has a payment order, logs in the module in the K-1 period but does not have the payment order, logs in the module in the K period and pays the order, and indicates that the vehicle owner is a paid return vehicle owner in the preferential refueling module in the K period, and the mark of the vehicle owner is 1; otherwise, the vehicle owner logs in the preferential fueling function module in the K-2 period and has a payment order, the vehicle owner logs in the module in the K-1 period but does not have the payment order, and the vehicle owner logs in the module in the K period but does not have the payment order, which indicates that the vehicle owner is the payment lost vehicle owner in the preferential fueling module in the K period and is marked as 0.

In the process of constructing training and testing samples, vehicle owner sample data is constructed by using vehicle owner characteristics, item characteristics and vehicle owner scene labels in the K-1 period, and the constructed vehicle owner sample data (training + testing) is distinguished to obtain sparse characteristics and dense characteristics. Wherein onehot processing is performed on sparse features, PCA decorrelation processing, normalization (standardization) processing, feature discretization processing and the like are performed on dense features. Randomly cutting the processed sparse features, the processed dense features and the scene labels of the car owners into training samples (the proportion is a) and testing samples (the proportion is 1-a) according to a certain proportion, for example, randomly cutting the samples into the training samples according to general experience: test sample =8:2 (i.e., training and test samples are randomly sliced at an 8:2 ratio).

When a prediction sample is constructed, constructing vehicle owner prediction sample data by using the user characteristic, the item characteristic and the vehicle owner scene label in the K period, distinguishing the prediction sample, and dividing the prediction sample into a sparse characteristic and a dense characteristic. Wherein, the sparse type characteristic is subjected to onehot processing, and the dense type characteristic is subjected to PCA decorrelation processing, normalization (standardization) processing, characteristic discretization processing and the like.

Wherein, the owner's characteristics (features) of K-1 stage mainly include: basic attribute data such as gender, age, region and the like of the car owner; active attribute data such as active days, active duration, active function quantity, interval of registration time and current time days and the like; recharging attribute data such as recharging amount, consumption amount, recharging times, recharging days, interval between the first recharging and the current time days and the like; owner function clicks, owner pick-up gift bag/type of gift certificate (quantity, number, value), use gift bag/type of gift certificate (quantity, value), expired gift bag/type of gift certificate (quantity, value), etc.

And finally, outputting training samples, testing samples and prediction samples under each scene.

In the single scene model training and testing stage 802, training samples and testing samples of the five scenes (a pull-in scene, a loss early warning scene, a loss pull-back scene, a pay pull-in scene and a pay backflow scene) are input into an LR binary classification model, so that model training and testing are performed on the training samples and the testing samples of each scene. The method comprises the steps of testing an LR (low-rate) binary model by using a training sample, testing the LR binary model obtained by training by using the testing sample after reaching a test finishing condition, and respectively storing the model weight vector if an evaluation index (indexes such as recall ratio, precision ratio, AUC (total efficiency index) and the like) reaches an evaluation effect

，

，

，

，

. Wherein the content of the first and second substances,

a model weight vector representing a pull-new scene;

representing a model weight vector of a loss early warning scene;

a model weight vector representing an attrition recovery scenario;

a model weight vector representing a pay pull scene;

a model weight vector representing a paid reflow scenario.

In the single scene prediction scoring stage 803, model weight vectors for each scene are input

，

，

，

，

And predicting the prediction samples corresponding to all scenes by adopting an LR algorithm to obtain probability score vectors under all scenes

，

，

，

. Wherein the content of the first and second substances,

a probability score vector representing a pull-new scene;

a probability score vector representing a loss early warning scenario;

a probability score vector representing an attrition recovery scenario;

a probability score vector representing a pay-pull new scene;

a probability score vector representing a paid reflow scenario.

In the single scene car owner classification phase 804, a threshold is given for each scene

，

，

，

，

Then the probability score vectors of five scenes obtained in the single scene prediction scoring stage 803 are used

，

Each owner probability score in (1) is compared with a corresponding scene threshold, if the owner probability score is greater than the scene threshold, the owner probability score is marked as 1, otherwise, the owner probability score is marked as 0. For example, for an owner probability score in scenario 1

Is greater than

If so, the owner is marked as 1, which indicates that the owner is the new owner (target owner) in the scene 1. The owner set of the target vehicle under each scene is obtained in the stage

，

，

，

，

. Wherein the content of the first and second substances,

representing the owner set of the target vehicle pulling the new scene,

a probability score vector representing a class of pull-new scenes as 1;

a target vehicle owner set representing a runoff early warning scenario,

a probability score vector representing that the loss early warning scene is classified as 1;

a target vehicle owner set representing a churn retrieval scenario,

a probability score vector representing a churn retrieval scenario classification of 1;

representing the target owner set for the pay-to-pull scenario,

a probability score vector representing a classification of the pay-pull scene as 1;

a target owner set representing a paid reflow scenario,

representing a paid reflow scenario classification ofA probability score vector of 1.

In the multi-scene fusion matrix construction stage 805, a collaborative filtering matrix with missing data is constructed based on the target vehicle owners of each scene obtained in the single-scene vehicle owner classification stage 804. Each row of the matrix represents the score vector of each target owner in five scenes, each column represents the score vectors of all the target owners in the scene, and the missing value represents that some users have no score or appear in some scenes or are not target users of the scene.

FIG. 9 is a schematic diagram of a collaborative filtering matrix according to an embodiment of the present application, and as shown in FIG. 9, users look at the rowsu ₁For the target users of scene 1, scene 2 and scene 5,u ₂target users for scene 1, scene 2, scene 3, scene 4 and scene 5,u ₃for the target users of scene 3 and scene 4,u ₄for the target users of scene 1, scene 2, scene 4 and scene 5,u ₅for the target users of scene 2, scene 4 and scene 5,u ₆the target users of scene 1, scene 3 and scene 4 correspond to the target users of scene 1 in terms of the number of columnsu ₁、u ₂、u ₄Andu ₆(ii) a Target users of scene 2 haveu ₁、u ₂、u ₄Andu ₅(ii) a Target users of scene 3 haveu ₂、u ₃Andu ₆(ii) a Target users of scene 4 haveu ₁、u ₂ 、u ₃、u ₄、u ₅Andu ₆(ii) a Target users of scene 5 haveu ₂、u ₄Andu ₅。

as can be seen from fig. 9, the collaborative filtering matrix at this time is with missing values.

In the collaborative filtering multi-scenario model prediction stage 807, the collaborative filtering model (CF) is used to predict and fill missing values of the collaborative filtering matrix with missing values, so as to obtain a filled score matrix as shown in fig. 10.

In the matrix, circles filled with oblique lines indicate predicted values obtained by the CF model. Therefore, a scoring matrix of all target car owners in each scene is obtained.

In the multi-scenario score sorting stage 808, based on the score matrix obtained by the collaborative filtering multi-scenario model prediction stage 807 after being filled by the CF model, each target user is sorted from large to small in each scenario according to the scenario score to obtain the score sorting vector of each owner in five scenarios, that is:

wherein, in the step (A),

representing the ith user, the highest ranked scene score, and so on.

And a scene recommending stage 809, selecting a scene with the highest score ranking for each vehicle owner to recommend the marketing campaign based on the ranked score vectors of the vehicle owners, so as to obtain an optimal marketing campaign scene scheme for each vehicle owner.

In the recommendation method provided by the embodiment of the application, the problem of independent marketing activities of each user due to different product life cycles of each user is solved by means of fusion of marketing scenes, cost waste can be effectively reduced, accurate recommendation is achieved, disturbance of marketing recommendation activities to the users is effectively reduced, marketing activity experience of the users is improved, marketing recommendation effects are improved in a targeted mode, the problem of effect superposition of multiple scene marketing activities of the users due to the fact that the users have multiple scenes in the related technology can be solved, the life cycle stage of the products where the users are located can be effectively determined, and the most effective recommendation scheme for the users can be effectively analyzed.

Continuing with the exemplary structure of the recommendation device 454 embodied as software modules provided by the embodiments of the present application, in some embodiments, as shown in fig. 3, the software modules stored in the recommendation device 454 in the memory 450 may include:

a first obtaining module 4541, configured to obtain, when it is determined that a recommendation opportunity is reached, user feature data corresponding to a plurality of trained recommendation models and a plurality of user identifiers respectively corresponding to a plurality of marketing scenario types;

the prediction module 4542 is configured to input the user feature data into each recommendation model, so as to obtain a prediction vector corresponding to each marketing scenario type;

the scene fusion module 4543 is configured to construct a multi-scene fusion matrix based on the prediction vector corresponding to each marketing scene type, where the multi-scene fusion matrix includes prediction probabilities of a plurality of target user identifiers in each marketing scene type;

a first determining module 4544, configured to determine, based on the multi-scenario fusion matrix, a target marketing scenario type corresponding to each target user identifier;

and the first sending module 4545 is configured to send recommendation information corresponding to the target marketing scenario type to a terminal corresponding to each target user identifier.

In some embodiments, the scene fusion module is further configured to:

In some embodiments, the first determining module is further configured to:

In some embodiments, the apparatus further comprises:

In some embodiments, the second obtaining module is further configured to:

In some embodiments, the first training module is further configured to:

In some embodiments, the apparatus further comprises:

Here, it should be noted that: the above description of the recommender embodiment is similar to the method description above with the same benefits as the method embodiment. For technical details not disclosed in the embodiments of the apparatus proposed by the present application, a person skilled in the art shall refer to the description of the embodiments of the method of the present application for understanding.

Embodiments of the present application provide a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to execute the recommendation method of the embodiment of the present application.

Embodiments of the present application provide a computer-readable storage medium storing executable instructions, which when executed by a processor, cause the processor to perform a recommendation method provided by embodiments of the present application, for example, a recommendation method as shown in fig. 4 and 6.

In some embodiments, the computer-readable storage medium may be memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash, magnetic surface memory, optical disk, or CD-ROM; or may be various devices including one or any combination of the above memories.

In some embodiments, executable instructions may be written in any form of programming language (including compiled or interpreted languages), in the form of programs, software modules, scripts or code, and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

By way of example, executable instructions may correspond, but do not necessarily have to correspond, to files in a file system, and may be stored in a portion of a file that holds other programs or data, such as in one or more scripts in a hypertext Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code).

By way of example, executable instructions may be deployed to be executed on one computing device or on multiple computing devices at one site or distributed across multiple sites and interconnected by a communication network.

The above description is only an example of the present application, and is not intended to limit the scope of the present application. Any modification, equivalent replacement, and improvement made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. A recommendation method, characterized in that the method comprises:

wherein the targeted marketing scenario type comprises at least one of: a pull-in scene, a loss early warning scene, a loss pull-in scene, a pay-for-new scene and a pay-for-return scene;

the pull-up scene is a marketing scene that the user has not logged in before the Kth time period;

the loss early warning scene is a marketing scene that a user logs in the (K-1) th time period and does not log in the K time period;

the loss recovery scene is a marketing scene that a user logs in the (K-2) th time period, does not log in the (K-1) th time period and logs in the K th time period;

the newly added pay scene is a marketing scene in which the user logs in but does not consume before the Kth time period and logs in and consumes in the Kth time period;

the paying backflow scene is a marketing scene that a user logs in and consumes in the (K-2) th time period, logs in and does not consume in the (K-1) th time period, and logs in and consumes in the K th time period;

2. The method of claim 1, wherein the constructing a multi-scenario fusion matrix based on the prediction vectors corresponding to the respective marketing scenario types comprises:

3. The method of claim 2, wherein constructing a multi-scenario fusion matrix based on the predicted probabilities of the respective target user identifications under at least one marketing scenario type comprises:

4. The method of claim 1, wherein determining the targeted marketing scenario type corresponding to each targeted user identifier based on the multi-scenario fusion matrix comprises:

5. The method of claim 1, further comprising:

when the preset recommendation interval time is determined to be reached, determining the recommendation time; alternatively, the first and second electrodes may be,

when the recommendation information is determined to be updated, determining that the recommendation time is reached; alternatively, the first and second electrodes may be,

and when the user is determined to log in, determining that the recommendation opportunity is reached.

6. The method according to any one of claims 1 to 5, further comprising:

obtaining log data corresponding to each user identification, and determining training data based on the log data, wherein the training data comprises a plurality of training characteristic data and marketing scene labels corresponding to each training characteristic data;

acquiring training characteristic data corresponding to each marketing scene type and a preset recommendation model corresponding to each marketing scene type;

and training the preset recommendation models corresponding to the marketing scene types respectively by using the training characteristic data corresponding to the marketing scene types to obtain the trained recommendation models corresponding to the marketing scene types.

7. The method of claim 6, wherein determining training data based on the log data comprises:

8. The method of claim 7, wherein determining historical feature data and marketing scenario tags corresponding to each user identifier based on the log data of each user identifier comprises:

9. The method of claim 7, wherein the training the preset recommendation model corresponding to each marketing scenario type by using the training feature data corresponding to each marketing scenario type to obtain the trained recommendation model corresponding to each marketing scenario type comprises:

10. The method of claim 9, further comprising:

when it is determined that a target recommendation model which does not reach a training end condition exists based on the test scene label and the prediction scene information, new training data corresponding to the target recommendation model is obtained again;

and continuously training the target recommendation model by using the new training data until a training end condition is reached, so as to obtain a trained recommendation model.

11. A recommendation device, characterized in that the device comprises:

the first determining module is used for determining a target marketing scene type corresponding to each target user identification based on the multi-scene fusion matrix; wherein the targeted marketing scenario type comprises at least one of: a pull-in scene, a loss early warning scene, a loss pull-in scene, a pay-for-new scene and a pay-for-return scene; the pull-up scene is a marketing scene that the user has not logged in before the Kth time period; the loss early warning scene is a marketing scene that a user logs in the (K-1) th time period and does not log in the K time period; the loss recovery scene is a marketing scene that a user logs in the (K-2) th time period, does not log in the (K-1) th time period and logs in the K th time period; the newly added pay scene is a marketing scene in which the user logs in but does not consume before the Kth time period and logs in and consumes in the Kth time period; the paying backflow scene is a marketing scene that a user logs in and consumes in the (K-2) th time period, logs in and does not consume in the (K-1) th time period, and logs in and consumes in the K th time period;

12. A recommendation device, characterized in that the recommendation device comprises:

a memory for storing executable instructions;

a processor for implementing the method of any one of claims 1 to 10 when executing executable instructions stored in the memory.

13. A computer-readable storage medium storing executable instructions, wherein the executable instructions, when executed by a processor, implement the method of any one of claims 1 to 10.